recursive vs. direct multi-step ahead forecasting

[ac_1]

Hence,

1. If I want both Direct Forecasts & Direct forecast SE's, and plot a fan chart, the LINREG calculations are applicable just for a stationary series.

2. If I am modelling a non-stationary I(1) series y, I can input the DIFFERENCES dy in LINREG, Direct Forecast the DIFFERENCES dyhat, back-out the Direct Forecast LEVELS (via accumulate dyhat and adding back to the last actual LEVEL y) but just for the DIRECT Forecasts?

[TomDoan]

LINREG can be used for non-stationary series.